2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer.h,v 1.100 2008/07/02 21:57:54 dillon Exp $
37 * This header file contains structures used internally by the HAMMERFS
38 * implementation. See hammer_disk.h for on-disk structures.
41 #include <sys/param.h>
42 #include <sys/types.h>
43 #include <sys/kernel.h>
45 #include <sys/systm.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mountctl.h>
50 #include <sys/vnode.h>
53 #include <sys/globaldata.h>
54 #include <sys/lockf.h>
56 #include <sys/queue.h>
58 #include <sys/globaldata.h>
61 #include <sys/signal2.h>
62 #include "hammer_disk.h"
63 #include "hammer_mount.h"
64 #include "hammer_ioctl.h"
66 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
68 MALLOC_DECLARE(M_HAMMER);
73 #if !defined(KTR_HAMMER)
74 #define KTR_HAMMER KTR_ALL
76 KTR_INFO_MASTER_EXTERN(hammer);
84 * Key structure used for custom RB tree inode lookups. This prototypes
85 * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
87 typedef struct hammer_inode_info {
88 int64_t obj_id; /* (key) object identifier */
89 hammer_tid_t obj_asof; /* (key) snapshot transid or 0 */
90 u_int32_t obj_localization; /* (key) pseudo-fs */
92 struct hammer_btree_leaf_elm *leaf;
94 } *hammer_inode_info_t;
96 typedef enum hammer_transaction_type {
100 } hammer_transaction_type_t;
103 * HAMMER Transaction tracking
105 struct hammer_transaction {
106 hammer_transaction_type_t type;
107 struct hammer_mount *hmp;
112 struct hammer_volume *rootvol;
115 typedef struct hammer_transaction *hammer_transaction_t;
121 int refs; /* active references delay writes */
122 int lockcount; /* lock count for exclusive/shared access */
124 int exwanted; /* number of threads waiting for ex lock */
125 struct thread *locktd;
129 hammer_islocked(struct hammer_lock *lock)
131 return(lock->lockcount != 0);
135 hammer_isactive(struct hammer_lock *lock)
137 return(lock->refs != 0);
141 hammer_islastref(struct hammer_lock *lock)
143 return(lock->refs == 1);
147 * Return if we specifically own the lock exclusively.
150 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
152 if (lock->lockcount > 0 && lock->locktd == td)
158 * Flush state, used by various structures
160 typedef enum hammer_inode_state {
164 } hammer_inode_state_t;
166 TAILQ_HEAD(hammer_record_list, hammer_record);
169 * Pseudo-filesystem extended data tracking
171 struct hammer_pfs_rb_tree;
172 struct hammer_pseudofs_inmem;
173 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
174 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
175 hammer_pfs_rb_compare, u_int32_t);
177 struct hammer_pseudofs_inmem {
178 RB_ENTRY(hammer_pseudofs_inmem) rb_node;
179 struct hammer_lock lock;
180 u_int32_t localization;
181 struct hammer_pseudofs_data pfsd;
184 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
187 * Cache object ids. A fixed number of objid cache structures are
188 * created to reserve object id's for newly created files in multiples
189 * of 100,000, localized to a particular directory, and recycled as
190 * needed. This allows parallel create operations in different
191 * directories to retain fairly localized object ids which in turn
192 * improves reblocking performance and layout.
194 #define OBJID_CACHE_SIZE 1024
195 #define OBJID_CACHE_BULK 100000
197 typedef struct hammer_objid_cache {
198 TAILQ_ENTRY(hammer_objid_cache) entry;
199 struct hammer_inode *dip;
200 hammer_tid_t next_tid;
202 } *hammer_objid_cache_t;
205 * Associate an inode with a B-Tree node to cache search start positions
207 typedef struct hammer_node_cache {
208 TAILQ_ENTRY(hammer_node_cache) entry;
209 struct hammer_node *node;
210 struct hammer_inode *ip;
211 } *hammer_node_cache_t;
213 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
216 * Structure used to represent an inode in-memory.
218 * The record and data associated with an inode may be out of sync with
219 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
222 * An inode may also hold a cache of unsynchronized records, used for
223 * database and directories only. Unsynchronized regular file data is
224 * stored in the buffer cache.
226 * NOTE: A file which is created and destroyed within the initial
227 * synchronization period can wind up not doing any disk I/O at all.
229 * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
231 struct hammer_ino_rb_tree;
233 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
234 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
235 hammer_ino_rb_compare, hammer_inode_info_t);
237 struct hammer_rec_rb_tree;
238 struct hammer_record;
239 RB_HEAD(hammer_rec_rb_tree, hammer_record);
240 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
241 hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
243 TAILQ_HEAD(hammer_node_list, hammer_node);
245 struct hammer_inode {
246 RB_ENTRY(hammer_inode) rb_node;
247 hammer_inode_state_t flush_state;
249 TAILQ_ENTRY(hammer_inode) flush_entry;
250 struct hammer_record_list target_list; /* target of dependant recs */
251 u_int64_t obj_id; /* (key) object identifier */
252 hammer_tid_t obj_asof; /* (key) snapshot or 0 */
253 u_int32_t obj_localization; /* (key) pseudo-fs */
254 struct hammer_mount *hmp;
255 hammer_objid_cache_t objid_cache;
257 int error; /* flush error */
258 int cursor_ip_refs; /* sanity */
262 hammer_pseudofs_inmem_t pfsm;
263 struct lockf advlock;
264 struct hammer_lock lock; /* sync copy interlock */
266 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */
267 struct hammer_inode_data ino_data; /* in-memory cache */
268 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */
269 struct hammer_node_cache cache[2]; /* search initiate cache */
272 * When a demark is created to synchronize an inode to
273 * disk, certain fields are copied so the front-end VOPs
274 * can continue to run in parallel with the synchronization
275 * occuring in the background.
277 int sync_flags; /* to-sync flags cache */
278 off_t sync_trunc_off; /* to-sync truncation */
279 off_t save_trunc_off; /* write optimization */
280 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
281 struct hammer_inode_data sync_ino_data; /* to-sync cache */
284 typedef struct hammer_inode *hammer_inode_t;
286 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data)
288 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */
289 /* (not including atime/mtime) */
290 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */
291 #define HAMMER_INODE_PFSD 0x0004 /* obj_asof set based on pfsd */
292 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */
293 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */
294 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */
295 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */
296 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */
297 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */
298 #define HAMMER_INODE_VHELD 0x0400 /* vnode held on sync */
299 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */
300 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */
301 #define HAMMER_INODE_REFLUSH 0x2000 /* pipelined flush during flush */
302 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */
303 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */
305 #define HAMMER_INODE_TRUNCATED 0x00010000
306 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/
307 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */
308 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */
309 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */
310 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
312 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY| \
313 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \
314 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \
315 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
317 #define HAMMER_INODE_MODMASK_NOXDIRTY \
318 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
320 #define HAMMER_FLUSH_GROUP_SIZE 64
322 #define HAMMER_FLUSH_SIGNAL 0x0001
323 #define HAMMER_FLUSH_RECURSION 0x0002
326 * Used by the inode reclaim code to pipeline reclaims and avoid
327 * blowing out kernel memory or letting the flusher get too far
330 struct hammer_reclaim {
331 TAILQ_ENTRY(hammer_reclaim) entry;
335 #define HAMMER_RECLAIM_FLUSH 2000
336 #define HAMMER_RECLAIM_WAIT 4000
339 * Structure used to represent an unsynchronized record in-memory. These
340 * records typically represent directory entries. Only non-historical
341 * records are kept in-memory.
343 * Records are organized as a per-inode RB-Tree. If the inode is not
344 * on disk then neither are any records and the in-memory record tree
345 * represents the entire contents of the inode. If the inode is on disk
346 * then the on-disk B-Tree is scanned in parallel with the in-memory
347 * RB-Tree to synthesize the current state of the file.
349 * Records are also used to enforce the ordering of directory create/delete
350 * operations. A new inode will not be flushed to disk unless its related
351 * directory entry is also being flushed at the same time. A directory entry
352 * will not be removed unless its related inode is also being removed at the
355 typedef enum hammer_record_type {
356 HAMMER_MEM_RECORD_GENERAL, /* misc record */
357 HAMMER_MEM_RECORD_INODE, /* inode record */
358 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */
359 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */
360 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */
361 } hammer_record_type_t;
363 struct hammer_record {
364 RB_ENTRY(hammer_record) rb_node;
365 TAILQ_ENTRY(hammer_record) target_entry;
366 hammer_inode_state_t flush_state;
368 hammer_record_type_t type;
369 struct hammer_lock lock;
370 struct hammer_reserve *resv;
371 struct hammer_inode *ip;
372 struct hammer_inode *target_ip;
373 struct hammer_btree_leaf_elm leaf;
374 union hammer_data_ondisk *data;
378 typedef struct hammer_record *hammer_record_t;
381 * Record flags. Note that FE can only be set by the frontend if the
382 * record has not been interlocked by the backend w/ BE.
384 #define HAMMER_RECF_ALLOCDATA 0x0001
385 #define HAMMER_RECF_ONRBTREE 0x0002
386 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */
387 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */
388 #define HAMMER_RECF_UNUSED0010 0x0010
389 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */
390 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */
391 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */
394 * In-memory structures representing on-disk structures.
396 struct hammer_volume;
397 struct hammer_buffer;
400 struct hammer_reserve;
402 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
403 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
404 RB_HEAD(hammer_nod_rb_tree, hammer_node);
405 RB_HEAD(hammer_und_rb_tree, hammer_undo);
406 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
408 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
409 hammer_vol_rb_compare, int32_t);
410 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
411 hammer_buf_rb_compare, hammer_off_t);
412 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
413 hammer_nod_rb_compare, hammer_off_t);
414 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
415 hammer_und_rb_compare, hammer_off_t);
416 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
417 hammer_res_rb_compare, hammer_off_t);
420 * IO management - embedded at the head of various in-memory structures
422 * VOLUME - hammer_volume containing meta-data
423 * META_BUFFER - hammer_buffer containing meta-data
424 * DATA_BUFFER - hammer_buffer containing pure-data
426 * Dirty volume headers and dirty meta-data buffers are locked until the
427 * flusher can sequence them out. Dirty pure-data buffers can be written.
428 * Clean buffers can be passively released.
430 typedef enum hammer_io_type {
431 HAMMER_STRUCTURE_VOLUME,
432 HAMMER_STRUCTURE_META_BUFFER,
433 HAMMER_STRUCTURE_UNDO_BUFFER,
434 HAMMER_STRUCTURE_DATA_BUFFER
437 union hammer_io_structure;
441 LIST_ENTRY(worklist) node;
444 TAILQ_HEAD(hammer_io_list, hammer_io);
445 typedef struct hammer_io_list *hammer_io_list_t;
448 struct worklist worklist;
449 struct hammer_lock lock;
450 enum hammer_io_type type;
451 struct hammer_mount *hmp;
452 TAILQ_ENTRY(hammer_io) mod_entry; /* list entry if modified */
453 hammer_io_list_t mod_list;
455 int64_t offset; /* zone-2 offset */
456 int bytes; /* buffer cache buffer size */
457 int loading; /* loading/unloading interlock */
460 u_int modified : 1; /* bp's data was modified */
461 u_int released : 1; /* bp released (w/ B_LOCKED set) */
462 u_int running : 1; /* bp write IO in progress */
463 u_int waiting : 1; /* someone is waiting on us */
464 u_int validated : 1; /* ondisk has been validated */
465 u_int waitdep : 1; /* flush waits for dependancies */
466 u_int recovered : 1; /* has recovery ref */
467 u_int waitmod : 1; /* waiting for modify_refs */
468 u_int reclaim : 1; /* reclaim requested */
469 u_int gencrc : 1; /* crc needs to be generated */
472 typedef struct hammer_io *hammer_io_t;
474 #define HAMMER_CLUSTER_SIZE (64 * 1024)
475 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
476 #undef HAMMER_CLUSTER_SIZE
477 #define HAMMER_CLUSTER_SIZE MAXBSIZE
479 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
482 * In-memory volume representing on-disk buffer
484 struct hammer_volume {
486 RB_ENTRY(hammer_volume) rb_node;
487 struct hammer_volume_ondisk *ondisk;
489 int64_t nblocks; /* note: special calculation for statfs */
490 int64_t buffer_base; /* base offset of buffer 0 */
491 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
492 hammer_off_t maxraw_off; /* Maximum raw offset for device */
498 typedef struct hammer_volume *hammer_volume_t;
501 * In-memory buffer (other then volume, super-cluster, or cluster),
502 * representing an on-disk buffer.
504 struct hammer_buffer {
506 RB_ENTRY(hammer_buffer) rb_node;
508 struct hammer_volume *volume;
509 hammer_off_t zoneX_offset;
510 hammer_off_t zone2_offset;
511 struct hammer_reserve *resv;
512 struct hammer_node_list clist;
515 typedef struct hammer_buffer *hammer_buffer_t;
518 * In-memory B-Tree node, representing an on-disk B-Tree node.
520 * This is a hang-on structure which is backed by a hammer_buffer,
521 * indexed by a hammer_cluster, and used for fine-grained locking of
522 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer
523 * can contain multiple nodes representing wildly disassociated portions
524 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
526 * This structure uses a cluster-relative index to reduce the number
527 * of layers required to access it, and also because all on-disk B-Tree
528 * references are cluster-relative offsets.
531 struct hammer_lock lock; /* node-by-node lock */
532 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */
533 RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */
534 hammer_off_t node_offset; /* full offset spec */
535 struct hammer_mount *hmp;
536 struct hammer_buffer *buffer; /* backing buffer */
537 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */
538 struct hammer_node_cache_list cache_list; /* passive caches */
540 int loading; /* load interlock */
543 #define HAMMER_NODE_DELETED 0x0001
544 #define HAMMER_NODE_FLUSH 0x0002
545 #define HAMMER_NODE_CRCGOOD 0x0004
546 #define HAMMER_NODE_NEEDSCRC 0x0008
547 #define HAMMER_NODE_NEEDSMIRROR 0x0010
549 typedef struct hammer_node *hammer_node_t;
552 * List of locked nodes.
554 struct hammer_node_locklist {
555 struct hammer_node_locklist *next;
559 typedef struct hammer_node_locklist *hammer_node_locklist_t;
563 * Common I/O management structure - embedded in in-memory structures
564 * which are backed by filesystem buffers.
566 union hammer_io_structure {
568 struct hammer_volume volume;
569 struct hammer_buffer buffer;
572 typedef union hammer_io_structure *hammer_io_structure_t;
575 * The reserve structure prevents the blockmap from allocating
576 * out of a reserved bigblock. Such reservations are used by
577 * the direct-write mechanism.
579 * The structure is also used to hold off on reallocations of
580 * big blocks from the freemap until flush dependancies have
583 struct hammer_reserve {
584 RB_ENTRY(hammer_reserve) rb_node;
585 TAILQ_ENTRY(hammer_reserve) delay_entry;
591 hammer_off_t zone_offset;
594 typedef struct hammer_reserve *hammer_reserve_t;
596 #define HAMMER_RESF_ONDELAY 0x0001
598 #include "hammer_cursor.h"
601 * The undo structure tracks recent undos to avoid laying down duplicate
602 * undos within a flush group, saving us a significant amount of overhead.
604 * This is strictly a heuristic.
606 #define HAMMER_MAX_UNDOS 1024
607 #define HAMMER_MAX_FLUSHERS 4
610 RB_ENTRY(hammer_undo) rb_node;
611 TAILQ_ENTRY(hammer_undo) lru_entry;
616 typedef struct hammer_undo *hammer_undo_t;
618 struct hammer_flusher_info;
620 struct hammer_flusher {
621 int signal; /* flusher thread sequencer */
622 int act; /* currently active flush group */
623 int done; /* set to act when complete */
624 int next; /* next flush group */
625 int group_lock; /* lock sequencing of the next flush */
626 int exiting; /* request master exit */
627 int count; /* number of slave flushers */
628 int running; /* number of slave flushers running */
629 thread_t td; /* master flusher thread */
630 hammer_tid_t tid; /* last flushed transaction id */
631 int finalize_want; /* serialize finalization */
632 struct hammer_lock finalize_lock; /* serialize finalization */
633 struct hammer_transaction trans; /* shared transaction */
634 struct hammer_flusher_info *info[HAMMER_MAX_FLUSHERS];
638 * Internal hammer mount data structure
640 struct hammer_mount {
642 /*struct vnode *rootvp;*/
643 struct hammer_ino_rb_tree rb_inos_root;
644 struct hammer_vol_rb_tree rb_vols_root;
645 struct hammer_nod_rb_tree rb_nods_root;
646 struct hammer_und_rb_tree rb_undo_root;
647 struct hammer_res_rb_tree rb_resv_root;
648 struct hammer_buf_rb_tree rb_bufs_root;
649 struct hammer_pfs_rb_tree rb_pfsm_root;
650 struct hammer_volume *rootvol;
651 struct hammer_base_elm root_btree_beg;
652 struct hammer_base_elm root_btree_end;
658 int masterid; /* -1 or 0-15 - clustering and mirroring */
659 int rsv_inodes; /* reserved space due to dirty inodes */
660 int rsv_databufs; /* reserved space due to dirty buffers */
661 int rsv_databytes; /* reserved space due to record data */
662 int rsv_recs; /* reserved space due to dirty records */
664 int count_newrecords;
666 int inode_reclaims; /* inodes pending reclaim by flusher */
667 int count_inodes; /* total number of inodes */
668 int count_iqueued; /* inodes queued to flusher */
670 struct hammer_flusher flusher;
672 u_int check_interrupt;
675 struct hammer_io_list volu_list; /* dirty undo buffers */
676 struct hammer_io_list undo_list; /* dirty undo buffers */
677 struct hammer_io_list data_list; /* dirty data buffers */
678 struct hammer_io_list alt_data_list; /* dirty data buffers */
679 struct hammer_io_list meta_list; /* dirty meta bufs */
680 struct hammer_io_list lose_list; /* loose buffers */
681 int locked_dirty_space; /* meta/volu count */
682 int io_running_space;
683 int objid_cache_count;
684 hammer_tid_t asof; /* snapshot mount */
685 hammer_off_t next_tid;
686 int64_t copy_stat_freebigblocks; /* number of free bigblocks */
688 u_int32_t namekey_iterator;
689 struct netexport export;
690 struct hammer_lock sync_lock;
691 struct hammer_lock free_lock;
692 struct hammer_lock undo_lock;
693 struct hammer_lock blkmap_lock;
694 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES];
695 struct hammer_undo undos[HAMMER_MAX_UNDOS];
697 TAILQ_HEAD(, hammer_undo) undo_lru_list;
698 TAILQ_HEAD(, hammer_inode) flush_list;
699 TAILQ_HEAD(, hammer_reserve) delay_list;
700 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
701 TAILQ_HEAD(, hammer_reclaim) reclaim_list;
704 typedef struct hammer_mount *hammer_mount_t;
706 #define HAMMER_MOUNT_UNUSED0001 0x0001
708 struct hammer_sync_info {
717 extern struct vop_ops hammer_vnode_vops;
718 extern struct vop_ops hammer_spec_vops;
719 extern struct vop_ops hammer_fifo_vops;
720 extern struct bio_ops hammer_bioops;
722 extern int hammer_debug_io;
723 extern int hammer_debug_general;
724 extern int hammer_debug_debug;
725 extern int hammer_debug_inode;
726 extern int hammer_debug_locks;
727 extern int hammer_debug_btree;
728 extern int hammer_debug_tid;
729 extern int hammer_debug_recover;
730 extern int hammer_debug_recover_faults;
731 extern int hammer_debug_cluster_enable;
732 extern int hammer_count_inodes;
733 extern int hammer_count_iqueued;
734 extern int hammer_count_reclaiming;
735 extern int hammer_count_records;
736 extern int hammer_count_record_datas;
737 extern int hammer_count_volumes;
738 extern int hammer_count_buffers;
739 extern int hammer_count_nodes;
740 extern int64_t hammer_stats_btree_lookups;
741 extern int64_t hammer_stats_btree_searches;
742 extern int64_t hammer_stats_btree_inserts;
743 extern int64_t hammer_stats_btree_deletes;
744 extern int64_t hammer_stats_btree_elements;
745 extern int64_t hammer_stats_btree_splits;
746 extern int64_t hammer_stats_btree_iterations;
747 extern int64_t hammer_stats_record_iterations;
748 extern int hammer_count_dirtybufspace;
749 extern int hammer_count_refedbufs;
750 extern int hammer_count_reservations;
751 extern int hammer_count_io_running_read;
752 extern int hammer_count_io_running_write;
753 extern int hammer_count_io_locked;
754 extern int hammer_limit_dirtybufspace;
755 extern int hammer_limit_iqueued;
756 extern int hammer_limit_recs;
757 extern int hammer_bio_count;
758 extern int hammer_verify_zone;
759 extern int hammer_write_mode;
760 extern int64_t hammer_contention_count;
762 int hammer_vop_inactive(struct vop_inactive_args *);
763 int hammer_vop_reclaim(struct vop_reclaim_args *);
764 int hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
765 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
766 hammer_inode_t dip, u_int64_t obj_id,
767 hammer_tid_t asof, u_int32_t localization,
768 int flags, int *errorp);
769 void hammer_scan_inode_snapshots(hammer_mount_t hmp,
770 hammer_inode_info_t iinfo,
771 int (*callback)(hammer_inode_t ip, void *data),
773 void hammer_put_inode(struct hammer_inode *ip);
774 void hammer_put_inode_ref(struct hammer_inode *ip);
775 void hammer_inode_waitreclaims(hammer_mount_t hmp);
777 int hammer_unload_volume(hammer_volume_t volume, void *data __unused);
778 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
780 int hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused);
781 int hammer_install_volume(hammer_mount_t hmp, const char *volname);
783 int hammer_ip_lookup(hammer_cursor_t cursor);
784 int hammer_ip_first(hammer_cursor_t cursor);
785 int hammer_ip_next(hammer_cursor_t cursor);
786 int hammer_ip_resolve_data(hammer_cursor_t cursor);
787 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
789 int hammer_delete_at_cursor(hammer_cursor_t cursor, int64_t *stat_bytes);
790 int hammer_ip_check_directory_empty(hammer_transaction_t trans,
792 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
793 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
796 hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
797 void hammer_flush_record_done(hammer_record_t record, int error);
798 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
799 void hammer_rel_mem_record(hammer_record_t record);
801 int hammer_cursor_up(hammer_cursor_t cursor);
802 int hammer_cursor_up_locked(hammer_cursor_t cursor);
803 int hammer_cursor_down(hammer_cursor_t cursor);
804 int hammer_cursor_upgrade(hammer_cursor_t cursor);
805 int hammer_cursor_upgrade_node(hammer_cursor_t cursor);
806 void hammer_cursor_downgrade(hammer_cursor_t cursor);
807 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
809 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
810 int hammer_lock_ex_try(struct hammer_lock *lock);
811 void hammer_lock_sh(struct hammer_lock *lock);
812 void hammer_lock_sh_lowpri(struct hammer_lock *lock);
813 int hammer_lock_sh_try(struct hammer_lock *lock);
814 int hammer_lock_upgrade(struct hammer_lock *lock);
815 void hammer_lock_downgrade(struct hammer_lock *lock);
816 void hammer_unlock(struct hammer_lock *lock);
817 void hammer_ref(struct hammer_lock *lock);
818 void hammer_unref(struct hammer_lock *lock);
820 void hammer_sync_lock_ex(hammer_transaction_t trans);
821 void hammer_sync_lock_sh(hammer_transaction_t trans);
822 int hammer_sync_lock_sh_try(hammer_transaction_t trans);
823 void hammer_sync_unlock(hammer_transaction_t trans);
825 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
826 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
827 void hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
828 u_int64_t hammer_timespec_to_time(struct timespec *ts);
829 hammer_tid_t hammer_str_to_tid(const char *str);
830 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip);
831 void hammer_clear_objid(hammer_inode_t dip);
832 void hammer_destroy_objid_cache(hammer_mount_t hmp);
834 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
836 void hammer_clear_undo_history(hammer_mount_t hmp);
837 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
838 int hammer_get_dtype(u_int8_t obj_type);
839 u_int8_t hammer_get_obj_type(enum vtype vtype);
840 int64_t hammer_directory_namekey(const void *name, int len);
841 int hammer_nohistory(hammer_inode_t ip);
843 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
844 hammer_node_cache_t cache, hammer_inode_t ip);
845 int hammer_reinit_cursor(hammer_cursor_t cursor);
846 void hammer_normalize_cursor(hammer_cursor_t cursor);
847 void hammer_done_cursor(hammer_cursor_t cursor);
848 void hammer_mem_done(hammer_cursor_t cursor);
850 int hammer_btree_lookup(hammer_cursor_t cursor);
851 int hammer_btree_first(hammer_cursor_t cursor);
852 int hammer_btree_last(hammer_cursor_t cursor);
853 int hammer_btree_extract(hammer_cursor_t cursor, int flags);
854 int hammer_btree_iterate(hammer_cursor_t cursor);
855 int hammer_btree_iterate_reverse(hammer_cursor_t cursor);
856 int hammer_btree_insert(hammer_cursor_t cursor,
857 hammer_btree_leaf_elm_t elm);
858 int hammer_btree_delete(hammer_cursor_t cursor);
859 int hammer_btree_mirror_propagate(hammer_transaction_t trans,
860 hammer_node_t node, int index,
861 hammer_tid_t mirror_tid);
863 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
864 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
865 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
866 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
868 int btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
869 hammer_btree_elm_t elm);
870 int hammer_btree_lock_children(hammer_cursor_t cursor,
871 struct hammer_node_locklist **locklistp);
872 void hammer_btree_unlock_children(struct hammer_node_locklist **locklistp);
873 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
874 hammer_node_t hammer_btree_get_parent(hammer_node_t node, int *parent_indexp,
875 int *errorp, int try_exclusive);
877 void hammer_print_btree_node(hammer_node_ondisk_t ondisk);
878 void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
880 void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
881 int *errorp, struct hammer_buffer **bufferp);
882 void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
883 int *errorp, struct hammer_buffer **bufferp);
884 void *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
885 int *errorp, struct hammer_buffer **bufferp);
886 void *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
887 int *errorp, struct hammer_buffer **bufferp);
889 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
891 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
892 int32_t vol_no, int *errorp);
893 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
894 int bytes, int isnew, int *errorp);
895 void hammer_del_buffers(hammer_mount_t hmp, hammer_off_t base_offset,
896 hammer_off_t zone2_offset, int bytes);
898 int hammer_ref_volume(hammer_volume_t volume);
899 int hammer_ref_buffer(hammer_buffer_t buffer);
900 void hammer_flush_buffer_nodes(hammer_buffer_t buffer);
902 void hammer_rel_volume(hammer_volume_t volume, int flush);
903 void hammer_rel_buffer(hammer_buffer_t buffer, int flush);
905 int hammer_vfs_export(struct mount *mp, int op,
906 const struct export_args *export);
907 hammer_node_t hammer_get_node(hammer_mount_t hmp, hammer_off_t node_offset,
908 int isnew, int *errorp);
909 void hammer_ref_node(hammer_node_t node);
910 hammer_node_t hammer_ref_node_safe(struct hammer_mount *hmp,
911 hammer_node_cache_t cache, int *errorp);
912 void hammer_rel_node(hammer_node_t node);
913 void hammer_delete_node(hammer_transaction_t trans,
915 void hammer_cache_node(hammer_node_cache_t cache,
917 void hammer_uncache_node(hammer_node_cache_t cache);
918 void hammer_flush_node(hammer_node_t node);
920 void hammer_dup_buffer(struct hammer_buffer **bufferp,
921 struct hammer_buffer *buffer);
922 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans, int *errorp);
923 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
924 u_int16_t rec_type, hammer_off_t *data_offsetp,
925 struct hammer_buffer **data_bufferp, int *errorp);
927 int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io,
928 hammer_off_t zone1_offset, void *base, int len);
930 void hammer_put_volume(struct hammer_volume *volume, int flush);
931 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
933 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
934 hammer_off_t owner, int *errorp);
935 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
936 hammer_off_t owner, int *errorp);
937 int hammer_checkspace(hammer_mount_t hmp);
938 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
939 int bytes, int *errorp);
940 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
941 int bytes, hammer_off_t *zone_offp, int *errorp);
942 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
943 hammer_reserve_t resv);
944 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
945 void hammer_blockmap_free(hammer_transaction_t trans,
946 hammer_off_t bmap_off, int bytes);
947 void hammer_blockmap_finalize(hammer_transaction_t trans,
948 hammer_off_t bmap_off, int bytes);
949 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
950 int *curp, int *errorp);
951 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
953 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
955 int64_t hammer_undo_used(hammer_transaction_t trans);
956 int64_t hammer_undo_space(hammer_transaction_t trans);
957 int64_t hammer_undo_max(hammer_mount_t hmp);
959 void hammer_start_transaction(struct hammer_transaction *trans,
960 struct hammer_mount *hmp);
961 void hammer_simple_transaction(struct hammer_transaction *trans,
962 struct hammer_mount *hmp);
963 void hammer_start_transaction_fls(struct hammer_transaction *trans,
964 struct hammer_mount *hmp);
965 void hammer_done_transaction(struct hammer_transaction *trans);
967 void hammer_modify_inode(hammer_inode_t ip, int flags);
968 void hammer_flush_inode(hammer_inode_t ip, int flags);
969 void hammer_flush_inode_done(hammer_inode_t ip);
970 void hammer_wait_inode(hammer_inode_t ip);
972 int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
973 struct ucred *cred, struct hammer_inode *dip,
974 int pseudofs, struct hammer_inode **ipp);
975 void hammer_rel_inode(hammer_inode_t ip, int flush);
976 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
977 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
979 int hammer_sync_inode(hammer_inode_t ip);
980 void hammer_test_inode(hammer_inode_t ip);
981 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
983 int hammer_ip_add_directory(struct hammer_transaction *trans,
984 hammer_inode_t dip, const char *name, int bytes,
986 int hammer_ip_del_directory(struct hammer_transaction *trans,
987 hammer_cursor_t cursor, hammer_inode_t dip,
989 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
990 void *data, int bytes, int *errorp);
991 int hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
992 int hammer_ip_add_record(struct hammer_transaction *trans,
993 hammer_record_t record);
994 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
995 int64_t ran_beg, int64_t ran_end, int truncating);
996 int hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
998 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
999 int64_t offset, void *data, int bytes);
1000 int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1001 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1002 int hammer_load_pseudofs(hammer_transaction_t trans, hammer_inode_t ip);
1003 int hammer_save_pseudofs(hammer_transaction_t trans, hammer_inode_t ip);
1004 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1005 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1006 struct ucred *cred);
1008 void hammer_io_init(hammer_io_t io, hammer_mount_t hmp,
1009 enum hammer_io_type type);
1010 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
1011 hammer_off_t limit);
1012 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1013 void hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1014 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1015 void hammer_io_flush(struct hammer_io *io);
1016 void hammer_io_waitdep(struct hammer_io *io);
1017 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident);
1018 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio);
1019 int hammer_io_direct_write(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf,
1021 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1022 void hammer_io_write_interlock(hammer_io_t io);
1023 void hammer_io_done_interlock(hammer_io_t io);
1024 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1025 void hammer_io_clear_modlist(struct hammer_io *io);
1026 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1027 void *base, int len);
1028 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1029 void *base, int len);
1030 void hammer_modify_volume_done(hammer_volume_t volume);
1031 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1033 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1034 struct hammer_ioc_reblock *reblock);
1035 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1036 struct hammer_ioc_prune *prune);
1037 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1038 struct hammer_ioc_mirror_rw *mirror);
1039 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1040 struct hammer_ioc_mirror_rw *mirror);
1041 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1042 struct hammer_ioc_pseudofs_rw *pfs);
1043 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1044 struct hammer_ioc_pseudofs_rw *pfs);
1046 int hammer_signal_check(hammer_mount_t hmp);
1048 void hammer_flusher_create(hammer_mount_t hmp);
1049 void hammer_flusher_destroy(hammer_mount_t hmp);
1050 void hammer_flusher_sync(hammer_mount_t hmp);
1051 void hammer_flusher_async(hammer_mount_t hmp);
1052 int hammer_flusher_meta_limit(hammer_mount_t hmp);
1053 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1054 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1056 int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol);
1057 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1058 hammer_volume_t root_volume, int final);
1060 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1061 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1062 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1064 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1065 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1066 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1067 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1068 void hkprintf(const char *ctl, ...);
1070 int hammer_blocksize(int64_t file_offset);
1071 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1075 static __inline void
1076 hammer_wait_mem_record(hammer_record_t record)
1078 hammer_wait_mem_record_ident(record, "hmmwai");
1081 static __inline void
1082 hammer_lock_ex(struct hammer_lock *lock)
1084 hammer_lock_ex_ident(lock, "hmrlck");
1088 * Indicate that a B-Tree node is being modified.
1090 static __inline void
1091 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1093 hammer_modify_buffer(trans, node->buffer, NULL, 0);
1096 static __inline void
1097 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1099 hammer_modify_buffer(trans, node->buffer,
1100 node->ondisk, sizeof(*node->ondisk));
1103 static __inline void
1104 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1105 void *base, int len)
1107 hammer_crc_t *crcptr;
1109 KKASSERT((char *)base >= (char *)node->ondisk &&
1110 (char *)base + len <=
1111 (char *)node->ondisk + sizeof(*node->ondisk));
1112 hammer_modify_buffer(trans, node->buffer, base, len);
1113 crcptr = &node->ondisk->crc;
1114 hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1115 --node->buffer->io.modify_refs; /* only want one ref */
1119 * Indicate that the specified modifications have been completed.
1121 * Do not try to generate the crc here, it's very expensive to do and a
1122 * sequence of insertions or deletions can result in many calls to this
1123 * function on the same node.
1125 static __inline void
1126 hammer_modify_node_done(hammer_node_t node)
1128 node->flags |= HAMMER_NODE_CRCGOOD;
1129 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1130 node->flags |= HAMMER_NODE_NEEDSCRC;
1131 node->buffer->io.gencrc = 1;
1132 hammer_ref_node(node);
1134 hammer_modify_buffer_done(node->buffer);
1137 #define hammer_modify_volume_field(trans, vol, field) \
1138 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1139 sizeof((vol)->ondisk->field))
1141 #define hammer_modify_node_field(trans, node, field) \
1142 hammer_modify_node(trans, node, &(node)->ondisk->field, \
1143 sizeof((node)->ondisk->field))